Aizii Learn · AI Search
← Back to Learn

How often do AI models refresh their brand knowledge

No vendor publishes a refresh schedule for brand knowledge, and no vendor publishes a fixed time for a brand fact to update. That is the direct answer, and any guide that gives you a number of days for ChatGPT, Claude, Gemini, or Perplexity to learn a new fact is guessing. What vendors do publish is narrower: how their search crawlers work, how long a robots.txt change takes to reach their search systems, and how Google handles recrawling and indexing. Those are the levers you control. The rest is a mix of live search, which can move after a recrawl, and training, which moves only when a newer model arrives, on a schedule no vendor publishes for brands.

The confusion comes from treating brand knowledge as one thing. It is at least three things, and each refreshes on its own path. Live search answers are built from pages a crawler fetched recently, and they can change after the next crawl. Search indexes are built from pages an engine crawled and stored, and they change when the engine recrawls a page, which Google says can take anywhere from several days to several months depending on how often its systems determine a page needs to be refreshed. Training knowledge is built into a model when it is trained, and it changes only when a newer model is released. A price change can show up in a search based answer quickly and in a training based answer much later, and both can be true at the same time for the same brand.

Where it goes wrong

Teams plan around a refresh time that does not exist, and the plan fails in predictable ways.

  • A launch plan assumes a fixed waiting time. The team ships a product, waits the number of days a blog post promised, and finds the assistant still shows the old catalog. No vendor promised that number, so the wait was never a plan.
  • A price change is judged by the wrong answer type. The team tests a training based answer, sees the old price, and concludes the fix failed, when the search based answer already shows the new price. The two answer types move on different paths, and testing only one misleads the team.
  • A robots.txt change is judged too soon. OpenAI says it can take about 24 hours from a robots.txt update for its search systems to adjust, and Perplexity says it may take up to 24 hours for its systems to reflect changes. A retest ten minutes after the change proves nothing.
  • A Google recrawl is treated as instant. Google says crawling can take anywhere from several days to several months, and that a request for indexing does not guarantee the page will appear in the index. A team that expects an overnight change will read a normal delay as a failure.
  • The sitemap lies about change dates. Every page carries the same new date, stamped when the file was generated, not when the page changed. Google says it uses the lastmod value only if it is consistently and verifiably accurate. A file that stamps every page with the same date teaches the crawler to ignore the signal on every page.
  • The correction sits behind a block. The updated page is live, but the search crawler cannot fetch it, so the index keeps the old copy. The team waits for a refresh that cannot happen until access is fixed.
  • Training is treated as a refresh button. The team files a feedback report and expects the model to learn the correction. A feedback report is not a training update, and a source fix does not retrain a model. The training based answer waits for a newer model, and no vendor publishes when that arrives for a given brand.

The steps

Step 1. Work out which kind of answer you are watching. Ask your test prompt in each assistant and save whether the answer shows citations or links. An answer with citations to live pages is search based, and it can move after a recrawl. An answer with no citations that reads like remembered knowledge is likely training based, and it will not move until a newer model arrives. Log the two types separately from now on, because they answer different questions about freshness.

Step 2. Publish the change where a crawler can read it. Put the new fact in plain text on the live page, and make the structured data match it. Google says important content should be available in textual form, and that structured data should match the visible text. A fact that lives only in an image, a script, or a login wall is not a fact a crawler can refresh. Keep the page at a stable address, since a moved page without a redirect forces every engine to discover the new address from scratch.

Step 3. Keep the change dates truthful. Set the lastmod value in your sitemap from the real date the page last changed, and change it only when the main content, structured data, or links on the page change. Do not bump the date for a cosmetic edit. Do not stamp every page with the generation date. A trustworthy lastmod is the signal that tells a crawler which pages are worth re-fetching, and a file that lies about it gets ignored.

Step 4. Confirm the search crawlers can reach the changed page. Check robots.txt for OAI-SearchBot, PerplexityBot, and Googlebot, and check your server logs for visits that return a success response. Check the firewall and the CDN the same way, since OpenAI recommends allowing requests from its published search crawler addresses. Wait about 24 hours after a robots.txt change before you retest OpenAI or Perplexity search, since both vendors say their search systems take about that long to adjust.

Step 5. Ask Google to recrawl the changed pages. Inspect the changed address with the URL Inspection tool in Search Console and request indexing. Google says indexing can take up to a week or two, that a request does not guarantee the page will appear in the index, and that for many updated pages the best choice is to submit a sitemap with the updated pages marked by lastmod. Use the request for a few important pages, and use the sitemap for the rest. Do not submit the same unchanged file repeatedly.

Step 6. Retest on a schedule you can keep. Ask the same saved prompts again, with the same wording, and record the answer, the citations, and whether the answer was search based or training based. Retest search based answers after a recrawl, and retest training based answers only after a new model release you can name. A log with two columns, one for each answer type, stops the team from reading a training delay as a failed fix. If a search based answer is still stale after the page was recrawled, check the citation again: the assistant may be reading a retailer or directory page that still carries the old fact.

Step 7. Fix the slow pages, not the clock. If a page is recrawled rarely, the cause is usually the page itself: it changes rarely, few pages link to it, or it sits deep in the site. Link to it from the category page and the homepage, keep its change dates truthful, and keep it in the sitemap. Those are the levers Google documents for discovery. There is no setting that buys a faster refresh for a single brand, and no vendor sells one through a documented control.

The machine-readable piece

This sitemap snippet shows two pages with truthful lastmod values. The format comes from the sitemaps protocol, and the lastmod value uses a placeholder shape: replace YYYY-MM-DD with the real date the page last changed. A single sitemap holds up to 50,000 URLs and no more than 50MB uncompressed, so a large catalog needs several sitemaps grouped under an index file.

<?xml version="1.0" encoding="UTF-8"?>
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">
  <url>
    <loc>https://[yourdomain.com]/products/example-coffee-beans</loc>
    <lastmod>YYYY-MM-DD</lastmod>
  </url>
  <url>
    <loc>https://[yourdomain.com]/collections/coffee</loc>
    <lastmod>YYYY-MM-DD</lastmod>
  </url>
</urlset>

This robots.txt snippet keeps the search crawlers open so a changed page can be re-fetched. OpenAI documents OAI-SearchBot and GPTBot as independent settings, and Perplexity documents PerplexityBot as its search crawler. A change to this file takes about 24 hours to reach the search systems of both vendors, so note the time you change it.

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Googlebot
Allow: /

What it looks like when it works

On the page, the change is plain to a human reader. The product page shows the new price, the new size, or the new availability in normal text, and the structured data says the same thing. The sitemap lists the page with the real date it changed. The server logs show the search crawlers visiting and returning success responses. An assistant asked the same question gives the new fact and cites the changed page. That is the search based answer moving. The training based answer may still show the old fact, and that is not a failure of the fix. It is a different kind of knowledge, waiting for a newer model.

Say you run a coffee roaster and you change the price of your house blend and add a new single origin bean. A shopper asks an assistant for your current house blend price, and it gives the old price, citing your product page. You update the price in plain text and in the markup, add the new bean page, and update the sitemap with the real change dates for both pages. You check that OAI-SearchBot, PerplexityBot, and Googlebot can fetch them, and you request indexing of both pages in Search Console. You wait about 24 hours after any robots.txt change, then retest. The search based answer shows the new price and cites the updated page, and the new bean appears when shoppers ask about it. The training based answer still shows the old price. You log both, and you stop retesting the training answer until a new model release arrives.

The fundamentals

These mechanics are shared across every engine and covered in full in their own guides:

What this guide does not cover

This guide covers how brand knowledge refreshes across live search, search indexes, and training, and what you can do to speed up the parts you control. It does not cover a fixed refresh schedule, because no vendor publishes one, and this guide does not invent one. It does not cover correcting a wrong fact that was never true, which is a different job with its own guide. It does not cover the monitoring work of catching stale answers early, or the full technical audit that checks every crawler and every page. Those are separate jobs, and this guide is the timing picture, not the whole program.

Check your own store now

Run the free scan. One click, no signup, and it shows whether AI crawlers can read your store today: aizii.ai/aeo-scan

Run the free scan

Sources

Note: links marked specific point to the page that supports the claim in this guide. Links marked general docs point to the vendor's general docs section for that topic, not to a page that proves every step. Check the vendor docs for the current path before relying on a menu name or click path in this guide.